AI Briefing#vision-language-model
The latest AI and developer news about #vision-language-model, with the original source and a short summary.
The latest AI and developer news about #vision-language-model, with the original source and a short summary.
Multimodal models developed more effectively when modalities were trained together from the beginning.
PaddlePaddle has released HPD-Parsing, a 1B document parsing model based on hierarchical parallel decoding, achieving the top score of 94.91% on OmniDocBench v1.6.
Microsoft has released Mage, a lightweight multimodal model family with 4B parameters.
NVIDIA has unveiled Cosmos 3 Edge, a 4B-parameter open world model for edge devices.
NVIDIA has released Cosmos 3 Edge, a 4B-parameter world model capable of real-time inference on edge devices.
A model called VICIS has been proposed that enables Vision-Language Models to infer visual concepts from example images and apply them.
From the emergence of DALL-E through VQ-VAE-based models to LG AI Research's L-Verse, we look at the evolution of Vision-Language models.
Following NLP's success, the vision-language field is also accelerating large-scale model training leveraging massive data and infrastructure.
We introduce ChartInstruct, a general-purpose chart understanding model that comprehends the visual information in charts and performs complex reasoning.
We introduce ChartInstruct, a general-purpose chart vision-language model that applies Instruction Tuning to maximize chart understanding and reasoning capabilities.
Google has updated Gemma 4's chat template to improve tool calling performance and now supports Flash Attention 4.
Moondream 3.1, a model that applies MoE architecture to boost efficiency and performance, has been released.
Robostral Navigate is an 8B-scale embodied AI model that autonomously navigates complex environments using only a single RGB camera.
AllenAI has unveiled MolmoMotion, a model that predicts the 3D trajectories of objects based on natural language instructions.
Vista, a model based on Qwen 3.5 that precisely predicts the coordinates of GUI elements, has been released.
Qwen3.7-Plus, a multimodal agent that unifies vision and language to perform GUI and CLI tasks, has been unveiled.
Qwen has released Q-Judger, a VLM model that automatically evaluates text-to-image generation quality.
Nvidia has unveiled 'LocateAnything,' a Vision-Language Grounding model that is 10x faster than Qwen3-VL through parallel box decoding.
NuExtract3, a 4B-scale vision-language model specialized in document understanding and structured information extraction, has been released.
Marlin, a 2B open-source VLM based on Qwen3.5-VL, has been released.
MiniCPM has released MiniCPM-V 4.6, unveiling a 1.3B vision-language model for mobile use.
Using the UEyes dataset, VLM gaze predictions were compared against real eye-tracking data.
Microsoft has unveiled World-R1's 3D-consistent video generation results.
DSO uses RL to optimize steering transformations to reduce bias in VLMs and LLMs.